AI will create more jobs than it destroys. The historical evidence — from the printing press to the PC to the digital camera — shows that when technology makes a task cheaper, demand for that task expands rather than collapses. The people who capture those new jobs are the ones who build skills early. Here's the economic principle behind that pattern, and what it means for your career right now.
What's in this article
- The 160-year-old law that predicts AI will expand work, not eliminate it — and why economists have seen this pattern before
- The historical proof — the printing press, the PC, and the digital camera: what actually happened to the jobs everyone said would disappear
- The jobs that don't exist yet — and why the people building skills now will be first in line for them
- The two groups already losing — it's not just the people ignoring AI
- What to do about it — why prompt engineering is the single skill that changes your outcome
The article is a 6-minute read. Skim the headlines, read what interests you.
The headline writes itself: "AI Will Take Your Job." It's on the cover of magazines. It's the subject of parliamentary inquiries. It's what your colleagues talk about over lunch.
But there's a 160-year-old economic observation that suggests the headline is wrong — not because AI won't change work, but because it will change it in the opposite direction from what most people expect.
It's called Jevons Paradox. And if you understand it, you'll see the next decade very differently from most of the people around you.
What Jevons Noticed in 1865
William Stanley Jevons was a British economist worried about coal. As steam engines became more efficient, his contemporaries assumed Britain would burn less of it. Jevons looked at the data and found the opposite: coal consumption went up. More efficient engines made steam power cheaper, so it spread to more places, more industries, more applications. Demand exploded.
His insight: when a resource becomes easier and cheaper to use, people use vastly more of it — creating demand that didn't previously exist.
Why This Changes Everything About the AI Jobs Debate
When a task becomes cheaper and faster to complete, the total volume of that task — and the adjacent tasks it unlocks — multiplies. Gutenberg's printing press eliminated the medieval scribe, then created publishing, editing, bookselling, and authorship at a scale no scribe could have imagined. The PC was supposed to replace typists and bookkeepers; instead it created software, IT, web design, and digital marketing — categories of work that didn't exist before. The digital camera destroyed Kodak's film labs, then made photography so cheap and accessible that Instagram created an influencer economy and more images are captured per minute today than in the entire 20th century combined. The fear is always the same. So is the outcome.
The Jobs That Don't Exist Yet
In 2004, "social media manager" wasn't a job title. By 2014, it was a department. In 2008, "data scientist" didn't exist. By 2012, Harvard Business Review called it the sexiest job of the century.
The honest answer to "what jobs will AI create?" is: we don't know yet. We can't. Just as no one in 1995 predicted the role of "influencer" or "community manager," we can't name what emerges when AI is woven into every industry at scale.
What we can predict — from the consistent historical pattern — is that the volume of work touching AI will expand rather than contract. And that the people positioned to capture it are the ones building the skills now, while most people are still debating whether any of this is real.
The Two Groups Already Falling Behind
Andrej Karpathy — former head of AI at Tesla, co-founder of OpenAI — put it plainly: "People who are not using LLMs are already losing."
Not will lose. Already losing.
But here's the part that gets less attention: thinking you're using AI isn't the same as using it well.
There's a sizeable and growing group paying for ChatGPT or Claude every month, pasting things in occasionally, getting mediocre results, and concluding that AI is overhyped. They're not wrong that their results are mediocre. They're wrong about why.
The output of an AI system is almost entirely determined by the quality of the instruction you give it. A weak prompt produces weak output. Paying for a more powerful model and feeding it poor instructions doesn't buy you better results — it just costs more.
The people losing aren't just the ones ignoring AI. They're also the ones using it with the wrong mental model — treating it like a search engine, expecting it to read their mind, and wondering why the results don't match the promise.
The Skill That Changes the Outcome
Across every domain where AI is delivering results — legal, finance, marketing, operations — the pattern is consistent: the people getting disproportionate value are the ones who've learned to communicate with it precisely.
This is prompt engineering. Not a coding skill. A communication skill. Knowing how to give an AI enough context, constraint, and direction to produce something actually useful. The learning curve is steeper than most people expect — which is exactly why the gap between those who've done it and those who haven't is already meaningful, and widening.
What Jevons Tells You to Do
The risk isn't being replaced by AI. The risk is being replaced by a person using AI well.
That person is already out there — producing better outputs in less time, doing work that previously required a team, briefing agents precisely and capturing gains that are supposed to be the prize of this moment.
Unit 1 of AgentTongue Academy is completely free — no account, no card. It covers the foundations of prompting that separate effective AI use from expensive mediocrity. The full course — £39, one-time — covers advanced techniques, AI agents, business applications, and prompt security across eight units.
Jevons was ultimately optimistic: efficiency creates more demand, not less. The question is who captures it.
Sources: Jevons, W.S. (1865). The Coal Question. Macmillan · Bessen, J. (2015). Learning by Doing. Yale University Press · US Bureau of Labor Statistics, 1980–2020 · Karpathy, A. — public statements, 2024 · LinkedIn Workforce Report Q4 2024 · WEF Future of Jobs 2025.